MCPCB vs regular boring old double layer with thermal via's

[pretmetled]

This might be a pretty stupid question, but I'll risk asking anyways...

At what wattage do you stop using regular PCBs and start using metal core?

For some prototypes I was thinking to use 0.8 mm double sided, open vias, HASL finish, and hope for the best. Does anyone have any experience with the thermal performance on that when compared to MCPCBs?

Also, for prototyping where do you get your MCPCBs done at a reasonable price?

[patrickhuang]

You can refer to CREE thermal note.

http://www.cree.com/products/pdf/XLamp_PCB_Thermal.pdf

[pretmetled]

Thanks! I already read this particular one, but maybe you know of more app notes like this one? Always useful to have a few sources to compare and all that...

[Steve K]

that's a good question. I sat in on a short training session by some application engineers from Texas Instruments. The session discussed switching power supplies, and mentioned how some of the new IC packages had heatsink pads on the bottom. This means that you have to dump the heat into the circuit board instead of putting it into a heatsink as you would do with a TO-220. We ran out of time, so I never got to ask about the details of how to dump a couple of watts into the middle of a board. Thermal vias are a popular technique, but what are the limits? Or how would you estimate the thermal resistance of the board? i.e. for a board that is xx by yy millimeters with 1 ounce power and ground layers, what would the thermal resistance be?

Maybe I can send the T.I. guys a quick e-mail and see if they have any recommendations.....

regards,
Steve K.

[pretmetled]

We ran out of time, so I never got to ask about the details of how to dump a couple of watts into the middle of a board. Thermal vias are a popular technique, but what are the limits? Or how would you estimate the thermal resistance of the board? i.e. for a board that is xx by yy millimeters with 1 ounce power and ground layers, what would the thermal resistance be?

Maybe I can send the T.I. guys a quick e-mail and see if they have any recommendations.....
.

Precisely! That's the sort of thing I'd love to know. Some rudimentary way to get a reasonable estimate of the thermal resistance would really help. Just so you can do a back-of-the-envelope calculation for your PCB. In the end you'll just have to have it fabbed and test the real thing, no question.

What I haven't found so far in the various app notes is the influence of surface finish. I am planning to use HASL finish purely because it is cheaper. But maybe ENIG gives better results, wouldn't know... Maybe there's nothing else for it but to have 2 board runs, same design, different surface finish and then run some tests.

Anyways, it would be interesting to hear some of the TI recommendations!

[Steve K]

I haven't pestered the TI guys yet, but have been poking thru the TI web site. There is a Power Management Guide that can be found here:
http://www.ti.com/lit/sg/slvt145k/slvt145k.pdf

page 91 discusses their "PowerPad" and how to use vias and copper layers to get the heat out. Nothing too new there, but there are links to other app notes on thermal management. Here's the "PowerPad Made Easy" app note:
http://www.ti.com/lit/an/slma004b/slma004b.pdf

Overall.... I'm not confident that T.I. has a lot to offer for higher power packages like 3 watt LEDs. I think the Cree app note is a more detailed treatment of the subject.

regards,
Steve K.

[patrickhuang]

Precisely! That's the sort of thing I'd love to know. Some rudimentary way to get a reasonable estimate of the thermal resistance would really help. Just so you can do a back-of-the-envelope calculation for your PCB. In the end you'll just have to have it fabbed and test the real thing, no question.

What I haven't found so far in the various app notes is the influence of surface finish. I am planning to use HASL finish purely because it is cheaper. But maybe ENIG gives better results, wouldn't know... Maybe there's nothing else for it but to have 2 board runs, same design, different surface finish and then run some tests.

Anyways, it would be interesting to hear some of the TI recommendations!

You are right, cheaper is very important. BTW, surface finish is very thin , I think It's hard to measure their difference of temperature, even by T3ster.

[patrickhuang]

You are right, cheaper is very important. BTW, surface finish is very thin , I think It's hard to measure their difference of temperature, even by T3ster.

[patrickhuang]

You can refer to CREE thermal note.

http://www.cree.com/products/pdf/XLamp_PCB_Thermal.pdf

You are right, cheaper is very important. BTW, surface finish is very thin , I think It's hard to measure their difference of temperature, even by T3ster.

[pretmetled]

You are right, cheaper is very important. BTW, surface finish is very thin , I think It's hard to measure their difference of temperature, even by T3ster.

Cheaper is definitely important. For me it means the difference between "I have the budget to get it done" and "too expensive, forget it!".

As for surface finish thickness ... and feel free to correct me if I'm wrong, but AFAIK it's not just the thickness that matters. HASL is a little thicker than ENIG, but as far as I understand it that's not going to be the big thing. ENIG is significantly flatter / smoother. And since contact surface finish plays a pretty big role in thermal conductivity I was wondering how HASL compared to ENIG. If the budget was unlimited I'd pick ENIG, do question. But if the difference between HASL vs ENIG for thermal interfacing is small, then it will be whatever is cheaper. Which is HASL.

[pretmetled]

I haven't pestered the TI guys yet, but have been poking thru the TI web site. There is a Power Management Guide that can be found here:
http://www.ti.com/lit/sg/slvt145k/slvt145k.pdf

page 91 discusses their "PowerPad" and how to use vias and copper layers to get the heat out. Nothing too new there, but there are links to other app notes on thermal management. Here's the "PowerPad Made Easy" app note:
http://www.ti.com/lit/an/slma004b/slma004b.pdf

Overall.... I'm not confident that T.I. has a lot to offer for higher power packages like 3 watt LEDs. I think the Cree app note is a more detailed treatment of the subject.

regards,
Steve K.

Thanks! I actually went through the Power Management Guide when I was looking for led driver ICs. I'll check it out again for the thermal management stuff. As for 3 Watt ... mmmh, dunno. They (TI) manage to get rid of 3000+ mW in chipscale packages with thermal pads just fine. It's not as if leds have the monopoly on thermal problems. But the Cree app note definitely has plenty of good stuff in it. As far as I can tell, the main reason to pick app notes by LED vendors is that leds seem to be less forgiving than silicon devices at the moment. Hopefully that will change in the years to come...

[Steve K]

Precisely! That's the sort of thing I'd love to know. Some rudimentary way to get a reasonable estimate of the thermal resistance would really help. Just so you can do a back-of-the-envelope calculation for your PCB. In the end you'll just have to have it fabbed and test the real thing, no question.

What I haven't found so far in the various app notes is the influence of surface finish. I am planning to use HASL finish purely because it is cheaper. But maybe ENIG gives better results, wouldn't know... Maybe there's nothing else for it but to have 2 board runs, same design, different surface finish and then run some tests.

Anyways, it would be interesting to hear some of the TI recommendations!

I got a reply from the TI rep... not a lot of new info, but he did mention this study of the ways to improve thermal performance for a SOT-223 package by using different size areas of copper on the top, middle, and bottom of the board. It also discusses ways to measure the junction temperature of a mosfet (and presumably other devices with p-n junctions).

http://www.ti.com/lit/an/snva036a/snva036a.pdf

He recommends a book: 'There is also a book by Tony Kordyban called “Hot Air Rises and Heat Sinks” '

So... no great breakthroughs in getting heat out of a board. Mostly just ways to fine tune what you are probably already doing.


regards,
Steve K.

[pretmetled]

Actually that appnote turned out to be pretty useful. As you say nothing groundbreaking, but just a slightly different angle on things which helps add to the big picture.

And the book tip was useful too! Based on the author's name I found this here link which has all sorts of fun stuff on the subject.

So thanks!

[Steve K]

that's a good series of articles on heatsinking and thermal management in general! Very easy to read, and ought to make you a little smarter too.

Steve K.

[pretmetled]

At the very least it was amusing to read. I like that guy's brand of realistic pessimism. XD

[SemiMan]

Simple answer is there is no simple answer.

It is all about how much of the heat transfer budget can be assigned to the PCB.

- How long does this need to last?
- How efficient does it need to be?
- What is the LM80 data for current you need it to run at?


For a short service life product where every last bit of efficiency is not important, then a good FR4 may be good at just about any current.
For a product where you need every last bit of efficiency, long life, and you are running the LED towards the upper end of its range, then you need a very good metal core board.
If you have limited heat sink capability, metal core may be essential to keep the temp down.
If you have lots of heat sink capability, FR4 may be just fine at any drive current.

Semiman

[dat2zip]

You may be interested in the web via calculator. With the calculator you can play around with board thickness, size of hole, plating thickness of the via and get Amp capacity and thermal resistance.

I've been doing a lot of R&D on this subject lately and I prefer to order PCBs as thin as possible. Many PCB fab shops will make the PCBs as thin as 0.2mm. The thermal resistance using multiple vias is as good as an MCPCB or better.

Where does FR4 PCB wattage run out and MCPCB become more attractive. It's when your PCB can not move sufficient heat to a large heat sink and the device you are trying to keep cool is no longer kept within the MFG specifications.

A general observation. Since thermal resistance is rated in C/W (degrees C/Watt) it makes sense that at 1W your transition across the PCB will be the straight thermal resistance. If for example that number is 40C/W then at 1W your device will be 40C above the heat sink it is mounted to.

It gets worse as wattage goes up. At 2W the temperature rise is now 2X 40C. At 3W the temperature rise is now 3X. You can see that as power increases so does the multiplier on the thermal resistance. Once you get to 10W and higher you need thermal resistances that are less than 5C/W and preferred numbers that are less than 1.

With a Cree XR-E the thermal pad is large enough to put down some 20+ vias. With some tiny LEDs you are lucky if you can get 5. The two factors that will get you are power (watts) and thermal pad size if you are talking LEDs. The smaller LEDs are going to be tough as you won't be able to get 10+ vias on the power pad.

One 18 mil via on 0.8mm thick FR4 is according to the calculator 49.3C/W. 10 of these vias would give you 4.93C/W. An XM-L LED adds another 1.5-2C/W. If we combine the PCB and LED that makes it approximately 6.93C/W or so. At 3A the thermal resistance drop to the backside of the PCB will be (10W * 6.93C/W) or approximately 69.3C rise assuming you can put 10 18mil vias under the thermal pad. If the LED die is rated for 150C then with an infinite heat sink attached the maximum ambient you can stay in specifications is 150C/69.3 = 80C ambient.

It may be worth the small additional cost and go thinner. The same setup with 0.2mm PCB is a temperate drop from the die to backside of 33.1C at 10W of power. (3.3V @ 3A ~ 10W). In this case the XM-L LED is part of the limiting factor if I remember correctly the thermal resistance is 2C/W. Actually looking it up it's 2.5C/W for the LED. The board will be 1.31C/W with 10 vias on a 0.2mm board and the LED adds another 2.5C/W.

Wayne